Language and cultural prompts influence large language models' responses to culturally anchored mental health attitudes
Two factorial experiments (2 prompt languages, Chinese or English, by 2 cultural framings, China or the United States) presenting ChatGPT-4o and DeepSeek-V3 with the same description of a friend showing signs of mild depression. The outcomes were the models' assessment of depression severity and how likely they were to recommend seeing a therapist.
Publisher
PEC Innovation (Elsevier)
Published
23 Sept 2026
Added
today
Key Findings
- For both models, English prompts produced higher depression ratings and higher therapist-referral likelihood than Chinese prompts for the same scenario
- A US cultural framing lowered ChatGPT-4o's depression assessment but raised DeepSeek-V3's assessment and referral likelihood
- The authors conclude that LLM mental-health assessments and recommendations depend on linguistic and cultural features of the prompt
Methodology Notes
Vignette experiment, two models, 2x2 design. Issue dated December 2026; OpenAlex publication date 2026-09-23 (Crossref DOI created 2026-09-23). ScienceDirect walled. Verified via the Crossref record (title, authors) and the OpenAlex abstract, which appeared after 09-24. Number of runs per cell not stated in the abstract.
Sources
PEC Innovation article (DOI)(opens in a new tab) (primary)
Archived snapshot (Wayback Machine)(opens in a new tab) — preserved against link rot
Authors
Mian Jia, Flora Yat Nga Lam, Xuanqi Wang
Tags
Cite This
APA
Mian Jia, Flora Yat Nga Lam, Xuanqi Wang. (2026). Language and cultural prompts influence large language models' responses to culturally anchored mental health attitudes. PEC Innovation (Elsevier). https://doi.org/10.1016/j.pecinn.2026.100505